What Determines Household Use of Financial Transaction ...
Transcript of What Determines Household Use of Financial Transaction ...
What Determines Household Use of Financial Transaction Services?
Ryan M. Goodstein
and
Sherrie L.W. Rhine
Federal Deposit Insurance Corporation*
October 17, 2013 FDIC’s 3rd Annual Consumer Research Symposium
*Disclaimer: The views expressed in this presentation are those of the authors and do not necessarily reflect the views of the FDIC
Introduction Share of U.S. Families with a Bank Account, by Year
1.00
0.95
0.90
0.85
0.80
1989 1992 1995 1998 2001 2004 2007 2010
Source: Survey of Consumer Finances; triennial data between 1989-2010
Bank account ownership has been growing in recent decades
So has use of nonbank Alternative Financial Services (AFS) such as check cashing and money orders
(IBISWorld 2012; Apgar and Herbert 2004)
As of 2011, 25 percent of US households used a nonbank AFS within the past year, and 4 out of 5 AFS users also had a bank account (FDIC 2012)
2
Introduction
Our goal: To better understand the determinants of household financial transaction services use
This is an important issue for consumers and public policy Use of bank accounts is supported by consumer protection laws
and regulations which may not apply to AFS
Integration into the financial mainstream has benefits for households, and may also yield positive externalities at the community level
Why is AFS use growing? Lack of access to bank branches? (Caskey 1994)
Need for liquidity? (FDIC 2012; Gross et al. 2012)
More transparent fee structures? (FI$CA 2013)
3
Introduction
Previous literature shows that unbanked rates are higher among households with less income, less education, black, Hispanic, etc.
e.g. FDIC 2012; Rhine and Greene 2006; Barr 2009, 2011; Hogarth and O’Donnell 1997; Kooce-Lewis, Swagler, and Burton 1996; Caskey 1994, 1997
These household characteristics also are associated with higher rates of AFS use (e.g. FDIC 2012; Gross et al. 2012)
Not clear from previous research the extent to which household characteristics pick up the effects of other unobserved factors
4
Introduction
Key contributions of our study:
1. We address some of the omitted variable issues from previous literature. Specifically, we control for:
the presence of bank branches and AFS providers in the household’s local area
other neighborhood attributes which may influence household decision making
2. We examine bank account ownership and use of nonbank Alternative Financial Transaction Services (AFTS) products in a unified framework
5
Introduction
Our focus is on financial transaction services Bank account ownership
AFTS: money orders, check cashing, and remittances
Distribution of Households by Financial Transaction Services Use
Nbr Obs
Proportion Does Not Use AFTS
Uses AFTS All
Has Bank Account
Bank Only
28,289
0.746
Bank + AFTS
6,850
0.181
35,139
0.927
No Bank Account
Cash Only
988
0.026
AFTS Only
1,792
0.047
2,780
0.073
All 29,277
0.772
8,642
0.228
37,919
1.000
Notes: Unweighted proportions, based on analysis sample.
6
Introduction
Preview of Results
Household socio-economic characteristics (e.g. income, education) are the most important determinants of bank account ownership
Other demographic characteristics (e.g. race, ethnicity) have a much more important influence on whether the household uses nonbank AFTS in addition to a bank account, rather than on bank account ownership alone
The presence of bank branches and AFTS providers within 5 miles of the household’s residence has relatively modest effects on household use of financial transaction services
Other neighborhood attributes have, at most, a minor effect
7
Outline
Introduction
Data and Model
Results
Conclusions
8
Data
Household level data are from 2011 FDIC Unbanked/Underbanked Supplement to CPS Large, nationally representative sample Restricted-access geographic identifiers allow us to merge in localized data
“Local Market” for Financial Services Defined as the 5 mile radius around the centroid of the HH’s census tract Bank branch location data: FDIC’s 2011 SOD AFTS location data: Census’ 2011 ZCBP (as in Bhutta 2012) For both bank branch and AFTS, we include
Binary indicator: = 1 if number of locations > 0, = 0 otherwise Concentration: Number of locations per 1k population
Neighborhood attributes From 2011 ACS five-year estimates, tract population characteristics County-level property and violent crime rates (FBI’s 2011 UCR) County-level voter turnout rates for 2008 presidential election (CQ press)
9
Empirical Model
We estimate two alternative models (for household i in census tract j):
Bank Account Ownership (Logit)
= 𝑓(𝑧𝑗, 𝑦𝑖, 𝑥𝑖, 𝑎𝑗, 𝑠)
FTS Bundles (Multinomial Logit)
= 𝑓𝑘(𝑧𝑗, 𝑦𝑖, 𝑥𝑖, 𝑎𝑗, 𝑠)
where k = 1 if “Bank Only” (i.e. banked and does not use AFTS) k = 2 if “Bank plus AFTS” (i.e. banked and uses AFTS) k = 3 if “AFTS Only” (i.e. unbanked and uses AFTS) k = 4 if “Cash Only” (i.e. unbanked and does not use AFTS)
P 𝐵𝑎𝑛𝑘 𝐴𝑐𝑐𝑜𝑢𝑛𝑡𝑖𝑗 = 1
P 𝐹𝑇𝑆𝑖𝑗 = 𝑘
10
Empirical Model
Control Variables:
zj: local market for financial services (indicators and per capita number of bank branches/AFTS)
yi : socio-economic characteristics of the household (income, employment, homeownership, education)
xi : other demographic characteristics of the household (age, race, ethnicity, nativity, HH structure)
aj : local neighborhood attributes (tract population characteristics; voter turnout; crime rates)
s : geographic fixed effects (state level)
11
Empirical Model
Note that we do not account for potential selection issues
For example, if banks and AFTS providers choose to locate in areas where demand for their services is high, the magnitude of these estimated effects on HH financial services choice will be biased upward in magnitude
This concern is mitigated somewhat by the fact we control for a wide range of local population characteristics in our empirical model
12
Outline
Introduction
Data and Model
Results
Conclusions
13
Results: P(Bank Account = 1) Estimates from “standard” model that controls only for HH
characteristics are similar to previous literature
Based on the fully specified model that includes controls for local financial services market and neighborhood attributes, we find:
Partial Effects of Bank Branch and AFTS on P(bank account = 1) effect of 1
Estimated SD inc. Category Variable Partial Effect from mean
Bank Branches At least 1 Bank Branch 0.011**
AFTS Providers
Bank Branches per capita
At least 1 AFTS
(0.005)
0.001
(0.002)
-0.003
[-0.0005]
AFTS per capita
(0.004)
-0.005
(0.008)
[-0.0005]
Notes: Partial effects based on estimates from underlying logit model. Specification includes full set of controls. N = 37919; Pseudo R2 = 0.35
Geographic access to bank branches and AFTS locations has a relatively small effect on household bank status
14
Results: P(Bank Account = 1)
Partial Effects of Neighborhood Attributes on P(bank account = 1) Effect of 1
SD inc. Estimated from
Category
Tract Population
Crime Rates
Voter Turnout
Variable
TrctPopShr LMI
TrctPopShr Age <= 34
TrctPopShr Black
TrctPopShr Hispanic
TrctPopShr FamHH SnglHd
TrctPopShr <= HS Diploma
County Property Crime
County Violent Crime
County Voter Turnout Rate
Partial Effect
-0.009
(0.011)
0.031**
(0.016)
-0.021**
(0.009)
0.002
(0.009)
-0.034*
(0.020)
-0.079***
(0.012)
-0.000
(0.000)
0.000
(0.001)
0.053***
(0.020)
mean
[-0.001]
[0.003]
[-0.004]
[0.000]
[-0.003]
[-0.014]
[-0.001]
[0.001]
[0.006]
Estimated effects of neighborhood attributes are generally quite small in terms of economic magnitude
One exception: local educational attainment
1 SD increase from the mean in the share of tract population with HS diploma or less (i.e. from 43 to 60 percent) reduces likelihood of being banked by 1.4 pp
Notes: Partial effects based on estimates from underlying logit model. Specification includes full set of controls. N = 37919; Pseudo R2 = 0.35
15
Results: P(Bank Account = 1) Partial Effects of Socio-economic Characteristics on P(bank account = 1)
Estimated Category Variable Partial Effect
HH Income Middle Income -0.013***
(0.004)
Moderate Income -0.028***
(0.004)
Low Income -0.075***
(0.004)
HH LF Status Part-Time -0.011***
(0.004)
Unemployed -0.053***
(0.005)
NILF -0.040***
(0.003)
HH Tenure Non-Homeowner -0.052***
(0.003)
HH Education No HS Diploma -0.090***
(0.004)
HS Diploma -0.050***
(0.003)
Some College -0.025***
(0.003) Notes: Partial effects based on estimates from underlying logit model. Specification includes full set of controls. N = 37919; Pseudo R2 = 0.35
Effects of socio-economic characteristics are large in magnitude
Low income HH 8 pp less likely to be banked (relative to high income)
HH with no HS diploma are 9 pp less likely to be banked (relative to college degree)
16
Results: P(Bank Account = 1) Partial Effects of Selected Other Demographic
Characteristics on P(bank account = 1)
Estimated Category Variable Partial Effect
HH Age Age 34 or less -0.006
(0.004)
Age 55 to 64 0.031***
(0.004)
Age 65 or older 0.062***
(0.003)
HH Race Black -0.037***
(0.005)
Asian 0.014**
(0.007)
Other -0.026***
(0.007)
HH Ethnicity Hispanic -0.035***
(0.005)
HH Nativity Non-Citizen -0.022***
(0.005) Notes: Partial effects based on estimates from underlying logit model. Specification includes full set of controls. N = 37919; Pseudo R2 = 0.35
Demographic effects are important but generally smaller in magnitude than socio-economic factors
17
Results: FTS Bundles
We now shift to the MNL model; dependent variable is the bundle of financial transaction services used (Bank Only; Bank plus AFTS; AFTS Only; Cash Only)
18
Results: FTS Bundles Partial Effects of Bank Branch and AFTS on use of FTS Bundles
Outcome
Bank AFTS Cash Category Variable Bank Only plus AFTS Only Only
Bank Branches At least 1 Bank Branch 0.017** -0.007 -0.007* -0.004
(0.008) (0.008) (0.004) (0.003)
Bank Branches per capita 0.004 -0.002 -0.000 -0.002
(0.003) (0.003) (0.002) (0.003)
[0.003] [-0.002] [-0.000] [-0.002]
AFTS Providers At least 1 AFTS -0.000 -0.003 0.002 0.000
(0.007) (0.006) (0.004) (0.003)
AFTS per capita -0.056** 0.047** 0.007 0.002
(0.025) (0.021) (0.007) (0.006)
[-0.005] [0.005] [0.001] [0.000]
Proportion of Sample in Category 0.75 0.18 0.05 0.03 Notes: Partial effects based on estimates from underlying logit model. Specification includes full set of controls. N = 37919; Pseudo R2 = 0.19. Number in brackets is estimated effect of a 1 SD increase from the mean.
As in bank account model, effects of Bank branch and AFTS are fairly small in magnitude
Also as in bank account model, effects of other neighborhood attributes are generally quite small (not shown)
19
Results: FTS Bundles Partial Effects of Selected Socio-Economic Characteristics on FTS Bundles
Outcome
Bank AFTS Cash Category Variable Bank Only plus AFTS Only Only
HH Income Middle Income -0.017** 0.004 0.010*** 0.002
(0.007) (0.007) (0.003) (0.003)
Moderate Income -0.045*** 0.018*** 0.019*** 0.008***
(0.007) (0.007) (0.003) (0.003)
Low Income -0.084*** 0.010 0.050*** 0.024***
(0.007) (0.007) (0.003) (0.003)
HH Tenure Non-Homeowner -0.116*** 0.065*** 0.036*** 0.016***
(0.005) (0.005) (0.002) (0.002)
HH Education No HS Diploma -0.124*** 0.035*** 0.062*** 0.027***
(0.008) (0.008) (0.004) (0.003)
HS Diploma -0.075*** 0.025*** 0.038*** 0.012***
(0.006) (0.006) (0.003) (0.002)
Some College -0.051*** 0.026*** 0.023*** 0.002
(0.006) (0.005) (0.002) (0.002)
Proportion of Sample in Category 0.75 0.18 0.05 0.03 Notes: Partial effects based on estimates from underlying logit model. Specification includes full set of controls. N = 37919; Pseudo R2 = 0.19
Estimated effects of socio-economic characteristics are large in magnitude
Households at lower levels of socio-economic status are relatively more likely to be AFTS Only or Cash Only (i.e. no bank account) and less likely to be in the Bank+AFTS group
20
Results: FTS Bundles Partial Effects of Selected Demographic Characteristics on use of FTS Bundles
Outcome
Bank AFTS Cash Category Variable Bank Only plus AFTS Only Only
HH Age Age 34 or less -0.004 -0.001 0.005 0.001
(0.006) (0.006) (0.003) (0.002)
Age 55 to 64 0.039*** -0.009 -0.021*** -0.010***
(0.006) (0.006) (0.003) (0.003)
Age 65 or older 0.126*** -0.065*** -0.047*** -0.015***
(0.006) (0.006) (0.002) (0.002)
HH Race Black -0.154*** 0.116*** 0.020*** 0.018***
(0.010) (0.009) (0.004) (0.003)
Asian 0.017 -0.001 -0.030*** 0.014**
(0.012) (0.011) (0.004) (0.006)
Other -0.089*** 0.064*** 0.017*** 0.009*
(0.014) (0.014) (0.006) (0.005)
HH Ethnicity Hispanic -0.078*** 0.044*** 0.017*** 0.018***
(0.009) (0.009) (0.004) (0.004)
HH Nativity Non-Citizen -0.074*** 0.050*** 0.012*** 0.013***
(0.010) (0.010) (0.004) (0.004)
Proportion of Sample in Category 0.75 0.18 0.05 0.03 Notes: Partial effects based on estimates from underlying logit model. Specification includes full set of controls. N = 37919; Pseudo R2 = 0.19
Estimated effects on demographic characteristics are large, and at least as important as socio-economic attributes (unlike the Bank Account model)
Racial and ethnic minority groups are relatively more likely to be Bank+AFTS and less likely to be AFTS Only or Cash Only 21
Results: Discussion
In summary, our results indicate that:
Socio-economic characteristics (e.g. income, education) are the most important determinants of bank account ownership
Other demographic characteristics (e.g. race, ethnicity) have a much more important influence on whether or not a household uses nonbank AFTS in addition to a bank account.
We explore this further by stratifying sample on income
Is there heterogeneity in the magnitude of effects?
22
Results: Stratify by HH Income Partial Effects of Selected HH Characteristics on Bank Status
Model and Dependent Variable
Sample
Logit: P(Bank Account)
Low Income High Income
HH Education No HS Diploma
HS Diploma
Some College
HH Age Age 34 or less
Age 55 to 64
Age 65 or older
HH Race Black
Asian
Other
HH Ethnicity Hispanic
-0.198***
(0.014)
-0.136***
(0.014)
-0.079***
(0.014)
-0.011*
(0.007)
0.062***
(0.009)
0.157***
(0.009)
-0.065***
(0.009)
0.026
(0.021)
-0.056***
(0.014)
-0.063***
(0.009)
-0.022***
(0.006)
-0.015***
(0.005)
-0.011**
(0.005)
0.004
(0.004)
0.002
(0.003)
0.005
(0.005)
-0.009**
(0.004)
-0.009*
(0.005)
Sample Size
Proportion of Sample in Category
14493
0.84
9009
0.996
Among low-income households, household characteristics are associated with large differences in bank ownership rates
Among high-income households, effects of race, ethnicity are quite small in magnitude
Implications:
For more marginal HH, demographic characteristics have some influence
But above some minimum threshold level of income and education, nearly all households have a bank account regardless of their demographic characteristics
23
Results: Stratify by HH Income Partial Effects of Selected HH Characteristics on FTS Bundles
Model and Dependent Variable
Sample
MNL: P(Bank
Low Income
Only)
High Income
MNL: P(Ba
Low Income
nk + AFTS)
High Income
HH Education No HS Diploma
HS Diploma
Some College
HH Age Age 34 or less
Age 55 to 64
Age 65 or older
HH Race Black
Asian
Other
HH Ethnicity Hispanic
-0.193***
(0.015)
-0.140***
(0.014)
-0.093***
(0.014)
-0.016*
(0.010)
0.061***
(0.011)
0.225***
(0.011)
-0.164***
(0.012)
0.093***
(0.025)
-0.101***
(0.020)
-0.074***
(0.014)
-0.111***
(0.019)
-0.049***
(0.010)
-0.035***
(0.009)
0.010
(0.011)
0.011
(0.009)
0.051***
(0.014)
-0.114***
(0.014)
-0.002
(4.284)
0.012
(7.156)
-0.062***
(0.015)
-0.010
(0.014)
-0.002
(0.012)
0.009
(0.013)
0.005
(0.009)
0.000
(0.010)
-0.063***
(0.011)
0.096***
(0.011)
-0.042*
(0.024)
0.046**
(0.018)
0.013
(0.013)
0.099***
(0.019)
0.041***
(0.010)
0.029***
(0.009)
-0.009
(0.011)
-0.010
(0.009)
-0.049***
(0.014)
0.108***
(0.014)
0.058
(1.950)
0.043
(3.120)
0.057***
(0.015)
Sample Size
Proportion of Sample in Category
14493
0.63
9009
0.86
14493
0.21
9009
0.13
Even among high-income households, certain HH characteristics have an important influence on HH choice of FTS bundles
Note: Estimates for AFTS Only and Cash Only are omitted from this table.
24
Results: Alternative Specifications
Prepaid/payroll cards We have ignored such cards in our analysis (use is fairly low in
our sample) But these cards are gaining in popularity and may be a viable
option for households considering how to manage their finances We tried two alternative specifications, treating use of
prepaid/payroll cards as equivalent to using: 1. bank account 2. AFTS
In both cases, findings are qualitatively similar
Other robustness checks included: Alt definitions of local area (e.g. 2 mi; 10 mi; 25 mi) Alt measures of financial service providers (e.g. nbr of
banks/AFTS locations) Alt treatment of money orders (or remittances), where use does
not disqualify household from being categorized as “Bank Only”
25
Results: Other Caveats
Our estimates indicate that geographic access to bank branches/AFTS has a relatively minor influence on HH choice of financial transaction services
May be due in part to measurement error, but several alternative measures yielded same qualitative finding
Other unobserved supply side factors may still be important (e.g. fee structures of bank accounts/AFTS; access to funds)
Our MNL model imposes assumption of IIA
This may be an issue, at least in theory. In practice – not clear.
In future work we will evaluate the extent to which this is an issue, and potentially use a specification that relaxes the IIA assumption
26
Outline
Introduction
Data and Model
Results
Conclusions
27
Conclusions
The presence of bank branches and AFTS providers within 5 miles of the household’s residence has relatively modest effects on household use of financial transaction services
Other neighborhood attributes have, at most, a minor effect
28
Conclusions
Household socio-economic characteristics (e.g. income, education) are the most important determinants of bank account ownership
Other demographic characteristics (e.g. race, ethnicity) have a more important influence on whether the household uses nonbank AFTS in addition to a bank account, rather than on bank account ownership alone
Given that we control for the local market for financial services, and for other neighborhood attributes, the specific mechanisms driving the finding that black and Hispanic households are more likely to use AFTS in combination with bank accounts are not obvious
29
- the end -
Thank you!
30
Appendix: Selected Descriptive Stats
For all characteristics that vary at the person-level, we use householder characteristics to represent the household
We drop roughly 15% of households that participated in the CPS supplement due to missing data (primarily due to missing geography; sample means change minimally) 37919 obs
Selected Descriptive Statistics
Category Variable ALL
By Bank Status
Banked Unbanked
By Financial Services Type
Bank Bank AFTS Cash Only plus AFTS Only Only
Bank Branches At least 1 Bank Branch 0.89 0.89 0.90 0.89 0.88 0.90 0.89
Bank Branches per capita (1k) 0.33 0.33 0.29 0.34 0.31 0.30 0.28
AFTS Providers At least 1 AFTS 0.66 0.65 0.74 0.64 0.68 0.75 0.72
AFTS per capita (1k) 0.05 0.05 0.07 0.05 0.06 0.08 0.07
Tract Population TrctPopShr LMI 0.42 0.41 0.54 0.40 0.46 0.55 0.54
TrctPopShr Age <= 34 0.46 0.46 0.50 0.45 0.48 0.50 0.50
TrctPopShr Black 0.11 0.11 0.24 0.09 0.16 0.24 0.23
TrctPopShr Hispanic 0.12 0.12 0.22 0.11 0.15 0.22 0.22
TrctPopShr FamHH Single Head 0.17 0.16 0.24 0.16 0.19 0.25 0.24
TrctPopShr HS Diploma or Less 0.43 0.42 0.55 0.41 0.46 0.55 0.54
Crime Rates County Property Crime 28.62 28.39 31.55 28.11 29.53 31.80 31.10
County Violent Crime 3.89 3.82 4.72 3.74 4.16 4.79 4.58
Voter Turnout County Voter Turnout Rate 0.60 0.61 0.57 0.61 0.59 0.57 0.57
Number of Observations 37919 35139 2780 28289 6850 1792 988
Notes: unweighted proportions, analysis sample. Crime rates are per 100k population. 31